Direct training of dynamic observation noise with UMarineNet

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Accurate uncertainty predictions are crucial to assess the reliability of a model, especially for neural networks. Part of this uncertainty is the observation noise, which is dynamic in our marine virtual sensor task. Typically, dynamic noise is not trained directly, but approximated through terms in the loss function. Unfortunately, this noise loss function needs to be scaled by a trade-off-parameter to achieve accurate uncertainties. In this paper we propose an upgrade to the existing architecture, which increases interpretability and introduces a novel direct training procedure for dynamic noise modelling. To that end, we train the point prediction model and the noise model separately. We present a new loss function that requires Monte Carlo runs of the model to directly train for the uncertainty prediction accuracy. In an experimental evaluation, we show that in most tested cases the uncertainty prediction is more accurate than the manually tuned trade-off-parameter. Because of the architectural changes we are able to analyze the importance of individual parts of the time series of our prediction.

Original languageEnglish
Title of host publicationArtificial Neural Networks and Machine Learning – ICANN 2018 - 27th International Conference on Artificial Neural Networks, 2018, Proceedings
EditorsVera Kurkova, Barbara Hammer, Yannis Manolopoulos, Lazaros Iliadis, Ilias Maglogiannis
Number of pages11
PublisherSpringer Verlag,
Publication date1 Jan 2018
Pages123-133
ISBN (Print)9783030014179
DOIs
Publication statusPublished - 1 Jan 2018
Externally publishedYes
Event27th International Conference on Artificial Neural Networks, ICANN 2018 - Rhodes, Greece
Duration: 4 Oct 20187 Oct 2018

Conference

Conference27th International Conference on Artificial Neural Networks, ICANN 2018
LandGreece
ByRhodes
Periode04/10/201807/10/2018
SeriesLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11139 LNCS
ISSN0302-9743

    Research areas

  • CNN, LSTM, Predictive uncertainty, Time series

ID: 223195961